DeepSeek

4.3

DeepSeek is a high-performing LLM from China with class-leading scores on code and mathematical reasoning, API pricing a fraction of its competitors, and published open weights.

Commercial use OK
Free tier

Try DeepSeek on the official siteUsage-based API pricing

Pricing: Usage-based API pricing
Free tier: Free tier
Commercial use: Permitted
Japanese UI: Not confirmed
Enterprise contract: Not confirmed

What it is good for

Developers optimising API cost
Code and mathematical reasoning workloads
Teams that want to self-host open weights

Strengths and trade-offs

Strengths

Strong on code generation and maths
API pricing far below comparable models
Weights published, so self-hosting is possible

Trade-offs

Data residency and regulatory risk come up in corporate review
Production use with customer data needs a DPA and a close look at the data path

Editors' review

High-performing open models from China — cheap API, strong on code and maths

DeepSeek scores near the top of its class on code generation and mathematical reasoning, and its API costs a fraction of comparable models. The weights are published, so the models can also be run yourself.

Individuals get web chat with a free tier; developers use the usage-based API. Weights can be downloaded from Hugging Face and similar and hosted privately.

Because the vendor is based in China, data residency and regulatory risk come up in corporate review. Production use involving customer data needs a data processing agreement and scrutiny of the request path before it is approved.

Verdict

A strong option for individual developers and researchers who want capability and low API cost. Clear it with legal before putting it in production.

Compiled from official documentation and public sources, with our own editorial judgement. We do not publish figures we cannot source.